Google launches Gemini 4 Carbon, tops Artificial Analysis Intelligence Index at $2 per million input tokens
Google released its latest frontier model Gemini 4 Carbon on October 28, focused on coding, agentic tasks and cybersecurity. Google says it ranks first on the Artificial Analysis Intelligence Index v5.0, with Claude Opus 5.5 and GPT-6 Astra behind, holds the top spot on both SEC-bench Pro for real vulnerability discovery and ExploitGym for exploit generation, scores 71.0 percent on CWE-bench for remediation, and gains 89 Elo to 1768 on WebDev Arena, ranking fourth. Pricing starts at $2 per million input tokens and $10 per million output, moving to $4 and $20 after the introductory period. The model is currently available only to a select group of cybersecurity defense companies and will expand to more partners through the Fairwind Program. Safety figures show a 0.0 percent successful cheating rate and 0.4 percent attempted cheating, with unsupported claims roughly halved versus the previous generation.